Open Jessespace opened 5 years ago
Random is good for training, and you may take frames from one file so many times that every frame shoud be taken times.So ,there is no data-wasting if you set enough steps/epochs.
Random is good for training, and you may take frames from one file so many times that every frame shoud be taken times.So ,there is no data-wasting if you set enough steps/epochs.
I see, thanks!
I have questions about the input of images.I have researched the code, found that just random crop 16 images from a video as input. And the other frames just throw away. Does the original paper mean this? Just my own doubts, hope to get answers, thanks!
for i in range(s_index, s_index + num_frames_per_clip): image_name = str(filename) + '/' + str(filenames[i]) img = Image.open(image_name) img_data = np.array(img) ret_arr.append(img_data)